基于机器学习的CT影像组学模型预测低黏附性胃癌  

Gastric poorly cohesive carcinoma predicted by machine learning based on CT radiomics

作　　者：付佳 汪翔 张爽[3] 金璿[4] 秦乃姗[1] 邱建星[1] 邹英华[2] FU Jia;WANG Xiang;ZHANG Shuang;JIN Xuan;QIN Naishan;QIU Jianxing;ZOU Yinghua(Department of Medical Imaging,Peking University First Hospital,Beijing 100034,China;Department of Interventional Radiology and Vascular Surgery,Peking University First Hospital,Beijing 100034,China;Department of Pathology,Peking University First Hospital,Beijing 100034,China;Department of Chemtoherapy,Peking University First Hospital,Beijing 100034,China)

机构地区：[1]北京大学第一医院医学影像科,北京100034 [2]北京大学第一医院介入血管外科,北京100034 [3]北京大学第一医院病理科,北京100034 [4]北京大学第一医院化疗科,北京100034

出　　处：《实用放射学杂志》2023年第9期1453-1456,1461,共5页Journal of Practical Radiology

摘　　要：目的 探讨基于机器学习的CT影像组学模型预测低黏附性胃癌的价值.方法 回顾性分析经术后病理确诊的 116 例胃癌患者的临床及术前CT资料,根据 WHO 分型分为低黏附性胃癌 30 例、非低黏附性胃癌 86 例.于CT静脉期轴位图像勾画三维感兴趣区体积,提取影像组学特征,按照计算机随机数分为训练集和测试集,采用最小绝对收缩与选择算子(LASSO)筛选影像组学特征.由 4 种(逻辑回归、随机森林、决策树和支持向量机)机器学习算法构建预测低黏附性胃癌的模型,采用受试者工作特征(ROC)曲线评估模型诊断效能.结果 低黏附性与非低黏附性胃癌的静脉期(P=0.009)和延迟期 CT值(P<0.001)有统计学意义.筛选出 8 个影像组学特征构建的 4 种机器学习模型,模型的准确度均大于 70%.准确度最高的模型在训练集为随机森林,准确度为 93.54%;在测试集为决策树模型,准确度为 82.61%.结论 基于机器学习的 CT 影像组学模型术前预测低黏附性胃癌具有可行性,为胃癌的亚型诊断提供辅助信息.Objective To explore the predictive value of machine learning based on CT radiomics in patient with gastric poorly cohesive carcinoma.Methods The clinical data and preoperative CT imaging of 116 patients with postoperative pathology confirmed gastric cancer were andyzed retrospectively.According to WHO classification,all patients were divided into two groups,including gastric poorly cohesive carcinoma group(n=30)and non-poorly cohesive carcinoma group(n=86).The volume of region of interest was delineated on the axial image of CT venous phase,and all radiomics features were further extracted.All data were divided into train-ing set and test set according to computer random numbers,and the least absolute shrinkage and selection operator(LASSO)were used to screen the radiomics features.Four machine learning models,including logistic regression,random forest,decision tree and support vector machine,was constructed,and the receiver operating characteristic(ROC)curve was performed to evaluate the diag-nostic efficiency of four models,respectively.Results There were significant differences in CT values of the venous phase(P=O.009)and delayed phase(P<0.001)between gastric poorly cohesive carcinoma and non-poorly cohesive carcinoma groups.The diagnostic efficiency of the four types of machine learning models were all higher than 70%.The random forest in the training set was achieved the highest accuracy(93.54%),while the decision tree model in the test set was achieved the highest accuracy(82.61%).Conclusion The machine learning based on CT radiomics model can distinguish gastric poorly cohesive carcinoma,which may provide supplementary informationinclinicalpractice.

关 键 词：计算机体层成像 胃癌 影像组学 

分 类 号：R814.42[医药卫生—影像医学与核医学] R735.2[医药卫生—放射医学] R445[医药卫生—临床医学]